Kernel optimization of LS-SVM based on damage detection for smart structures

The method of damage detection is an important issue related to the self-detecting damage function for smart structures. Based on smart structures´ nonlinear, parallel features, and the existed intrinsic flaws of conventional neural networks, research on Support Vector Machine (SVM) used to detect damages for smart structures has become one of main researches recently. Aimed at the key and difficult research problem on SVM - the selection and construction of kernel functions, a mixed kernel function used to Least Square Support Vector Machine (LS-SVM) is constructed through analyzing the existed kernel functions of LS-SVM. Based on damage detection for smart structures, the parameters of LS-SVM with the mixed kernel are optimized by Genetic Algorithms (GA), and the detecting results are compared with that of LS-SVM based on RBF kernel. The result shows that, the accuracy of damage detection based on LS-SVM with mixed kernel is higher than that based on LS-SVM with RBF kernel under the same condition. Compared with LS-SVM with RBF kernel, LS-SVM with mixed kernel possesses the better dissemination ability and stronger learning ability by absorbing the advantages of RBF kernel and polynomial kernel function.